Managing and optimising cloud services is one of the main challenges faced by industry and academia. A possible solution is resorting to self-management, as fostered by autonomic computing. However, the abstraction layer provided by cloud computing obfuscates several details of the provided services, which, in turn, hinders the effectiveness of autonomic managers. Data-driven approaches, particularly those relying on service clustering based on machine learning techniques, can assist the autonomic management and support decisions concerning, for example, the scheduling and deployment of services. One aspect that complicates this approach is that the information provided by the monitoring contains both continuous (e.g. CPU load) and categorical (e.g. VM instance type) data. Current approaches treat this problem in a heuristic fashion. This paper, instead, proposes an approach, which uses all kinds of data and learns in a data-driven fashion the similarities and resource usage patterns among the services. In particular, we use an unsupervised formulation of the Random Forest algorithm to calculate similarities and provide them as input to a clustering algorithm. For the sake of efficiency and meeting the dynamism requirement of autonomic clouds, our methodology consists of two steps: (i) off-line clustering and (ii) on-line prediction. Using datasets from real-world clouds, we demonstrate the superiority of our solution with respect to others and validate the accuracy of the on-line prediction. Moreover, to show the applicability of our approach, we devise a service scheduler that uses the notion of similarity among services and evaluate it in a cloud test-bed.
Generative AI on Enterprise Cloud with NiFi and Milvus
Service Clustering for Autonomic Clouds Using Random Forest
1. Service Clustering for Autonomic
Clouds Using Random Forest
Rafael Brundo Uriarte
IMT Lucca
Sotirios Tsaftaris Francesco Tiezzi
IMT Lucca University of Camerino
CCGrid - 7th May 2015 - Shenzhen, China
10. Existing Approaches
Solutions which handle mixed data types usually are not
scalable (e.g. HClustream)
Expert intervention is not feasible due to the dynamism
Distance Metric Learning Approaches require labelled data
or are computationally expensive.
Requirements and Existing Solutions Uriarte, Tsaftaris and Tiezzi 9/29
12. Random Forest
Mixed Features
Large Number of Features
Efficient and Scales Well
Easily Parallelizable
RF+PAM Uriarte, Tsaftaris and Tiezzi 11/29
13. Random Forest
Clustering with Random Forest
Originally Developed for Classification
On-Line Random Forest
Intrinsic Measure of Similarity
Clustering Algorithm (e.g. PAM)
RF+PAM Uriarte, Tsaftaris and Tiezzi 12/29
15. Problems
Similarity Matrix (Big Memory Footprint)
Re-cluster on Every New Observation
RF+PAM Uriarte, Tsaftaris and Tiezzi 14/29
16. Solution: RF+PAM
Off-line Training and On-line Prediction
Similarity Learning and Standard Clustering
RF+PAM Uriarte, Tsaftaris and Tiezzi 15/29
17. Solution: RF+PAM
Build Forest, Calculate Similarities, Cluster, Select
the medoids and Store the references of the leaves.
RF+PAM Uriarte, Tsaftaris and Tiezzi 16/29
18. Solution: RF+PAM
Parse service and Assign the cluster of the most
similar medoid to it.
RF+PAM Uriarte, Tsaftaris and Tiezzi 17/29
21. Cluster Quality
Clustering quality compared to 2 other
approaches (same dataset)
Better results in all criteria
Connectivity - Connectedness of the clusters
Dunn Index - Cluster density and Separation
Silhouette - Confidence in the assignment
Evaluation Uriarte, Tsaftaris and Tiezzi 20/29
22. On-line Prediction
On-Line vs Batch Mode
K-Fold Cross-Validation
Compared the Adjusted Rand Index (ARI) for 2
datasets:
Monitoring data of Google’s production
clouds - 12500 servers
Requests of a grid of the Dutch Universities
Research Testbed (DAS-2) - 200 servers
Evaluation Uriarte, Tsaftaris and Tiezzi 21/29
23. Results: ARI
K Google DAS-2
100 0.81 (0.32) 0.70 (0.23)
50 0.75 (0.19) 0.68 (0.17)
20 0.73 (0.09) 0.67 (0.11)
10 0.70 (0.06) 0.63 (0.09)
5 0.69 (0.05) 0.61 (0.07)
Evaluation Uriarte, Tsaftaris and Tiezzi 22/29
24. Use Case
Schedules according to the Dissimilarity
Similar services separated
Algorithms:
1. Random
2. Dissimilarity
3. Isolated
Evaluation Uriarte, Tsaftaris and Tiezzi 23/29
25. Use Case
9 VMs
Arrival Rates
Types of Service
Services’ SLA
Evaluation Uriarte, Tsaftaris and Tiezzi 24/29
28. Summary
We propose RF+PAM to alleviate the problem
of limited knowledge in AC
Validated RF+PAM with 3 Experiments
Scheduling Algorithm
Conclusions Uriarte, Tsaftaris and Tiezzi 27/29
29. Future Works
More Use Cases
Better Implementation
Conclusions Uriarte, Tsaftaris and Tiezzi 28/29
31. Prune Trees
Parsing is very fast and efficient
Prune requires analysis (time consuming)
Conclusions Uriarte, Tsaftaris and Tiezzi 29/29
32. Retraining
Ratio of predictions/training services (user defined):
Parallel training
Trade-off between updating/prediction
Other solutions:
Dissimilarity to Medoids
On-line Clustering (Current Limitations and
Prediction Speed)
Conclusions Uriarte, Tsaftaris and Tiezzi 29/29